Soitec’s AI wafer strategy: deposits, fixed prices and faster capacity expansion

AI data-center expansion is turning Photonics-SOI wafers into a strategic supply-chain bottleneck. Soitec’s response is not to immediately build another fab: it is securing demand first through multi-year capacity reservations, then adding output to existing facilities in France and Singapore. The company says Photonics-SOI revenue should more than double from slightly above $100 million in fiscal 2026, assuming no material disruption in the AI market. 1

The deal structure: a reservation with financial consequences

Soitec is finalizing capacity-reservation agreements with more than 10 photonics customers. Around 80% were expected to be signed within one to two weeks of the company’s late-August disclosure, with the remainder expected within about a month. 30

The reported terms have four important elements:

• A multi-year commitment: Each customer agrees to reserve a specified wafer volume over several years.
• Fixed pricing for the committed volume: The agreed allocation is priced in advance rather than left entirely exposed to spot-market conditions.
• Staged cash deposits: Customers pay deposits as they ramp production, providing Soitec with financial backing for the reserved capacity. 30
• A downside for under-ordering: A customer that buys its committed volume gets its deposit back; a customer that falls short forfeits the deposit. 30

The agreements do not create an unlimited fixed-price option. If a customer wants more wafers than its contracted allocation, that additional volume triggers a new pricing discussion. Customers must also provide inventory data, a safeguard intended to discourage them from reserving more supply than they need simply to keep it away from competitors. 30

The available disclosures do not state the aggregate wafer volume covered by the agreements or the size of each deposit. That distinction matters: the announcement establishes the commercial mechanism, but not the total value of the secured backlog.

Why Photonics-SOI matters to AI infrastructure

Photonics-SOI is the substrate used to make silicon-photonics components for optical data links. As AI systems move larger amounts of data between processors, memory and network equipment, optical interconnects are becoming an increasingly important part of data-center infrastructure. Soitec describes Photonics-SOI as a growth area driven by AI data-center demand, and it says sales doubled year over year in the first quarter of fiscal 2027. 137

Soitec’s negotiating leverage comes from its position in the specialized substrate market. UBS estimates that the company has roughly 95% share of the Photonics-SOI market, according to reporting on the agreements. 17 That concentration helps explain why customers are willing to commit cash and accept volume-based penalties to secure future supply.

The revenue outlook shows why the contracts matter. Soitec reported Photonics-SOI revenue above $100 million in fiscal 2026 and expects fiscal 2027 revenue to more than double that level. Management has described the implied figure above $200 million as a floor, but the outlook remains conditional on there being no material disruption in the AI market. 111

How Soitec is adding capacity without immediately building a new fab

Soitec says current capacity can cover this year and next year by combining several shorter-term measures rather than relying on a single large construction project. Those measures include:

1. Reallocating shared capacity. Underused production capacity can be shifted between businesses.
2. Improving yield and output. Process improvements can increase the number of usable wafers produced from existing equipment.
3. Installing additional tools. New equipment can be added to cleanroom space that already exists.
4. Using more than one production geography. Singapore has now been qualified for high-volume Photonics-SOI manufacturing, giving Soitec an additional source of supply beyond France. 1721

The Singapore qualification is particularly important because Soitec had produced Photonics-SOI only in France until several months before the August disclosure. The company is also considering fitting out an unequipped Singapore building rather than constructing a new fab from scratch; it expects a decision within six to 12 months. 17

For now, Soitec does not expect to need a new manufacturing plant until approximately 2029 and does not see an immediate need for a U.S. fab. 17 That timetable reflects a preference for faster, lower-risk capacity expansion while customer commitments clarify how durable AI-related demand will be.

What the agreements mean for Soitec and its customers

The structure shifts part of the expansion risk from Soitec to its customers. Soitec receives clearer demand visibility and staged financial support before committing to additional tools or facility fit-outs. Customers, in return, receive a defined allocation at a fixed price and greater assurance that capacity will be available during the ramp.

The trade-off is rigidity. Customers must forecast their needs accurately enough to meet the committed volume, or they risk losing the deposit. They also cannot automatically extend the fixed price to extra purchases. Inventory-data requirements add another obligation, but they help distinguish genuine demand from defensive capacity hoarding.

In practical terms, the contracts turn an uncertain spot-supply problem into a bilateral planning arrangement: customers finance part of the commitment, while Soitec gains a clearer basis for expanding supply.

A wider pattern in the AI semiconductor supply chain

Soitec’s approach fits a broader industry pattern in which long-term demand is being matched with long-term capacity investment. TSMC, for example, forecasts more than 80% annualized growth in CoWoS advanced-packaging capacity from 2022 to 2027 and said it planned nine phases of wafer-fab and advanced-packaging facilities in 2026. 43

That comparison should be treated carefully. The supplied evidence does not establish a reliable figure for Nvidia’s share of CoWoS demand, and the often-repeated claim of a “ninefold” expansion is not supported here. A separate industry estimate projects a CoWoS supply-demand gap of about 10% by the end of 2026, but that figure is a reported estimate rather than a formal TSMC forecast. 4655

The more defensible conclusion is narrower: AI supply constraints extend beyond leading-edge GPU fabrication. They can appear in optical substrates, advanced packaging and memory, where qualification requirements and lengthy capacity build times make short-term supply difficult to add. Deposits, prepayments and multi-year reservations give suppliers confidence to invest while giving buyers a way to secure scarce inputs.

The takeaway

Soitec is using its strong Photonics-SOI position to lock in demand before scaling aggressively. The agreements require specified multi-year volumes, staged deposits, inventory transparency and a financial penalty for under-ordering, while excess volume is subject to renegotiated pricing. Near-term growth will come from better yields, added equipment, internal capacity shifts and Singapore production—not an immediate greenfield fab.

The key uncertainty is not the existence of the reservation model but its scale: Soitec has disclosed the number of customers and the broad signing timetable, yet not the total wafer commitments or deposit amounts. Those details will determine how much of the projected Photonics-SOI growth is already commercially secured.